Bring your own keys

Create AI apps.
Ship anywhere.

Build the interface, the agents, the data and the permissions in one workspace — then run it for real against live APIs.

Open the Studio See how it works

Live tools

Real APIs, no mocks

Templates

Each runs end to end

Step kinds

Agents, tools, approval

Checkpointed

Every step, before the next

We never hold your keys

Your Anthropic key and every connector credential belong to you. They are stored in your own browser and sent straight to the provider on each call. Nothing is written to our database, because there is nothing of yours in it to write.

The practical consequence: you pay your provider directly, at cost, and you can revoke access to this tool at any time without asking us to do anything.

The workspace

Everything in one canvas

The page, the agents, the workflow and the data are the same document — so what you arrange is what runs.

The AOS Builder studio: a page canvas on the left, an inspector on the right.

What you get

A builder that has to survive contact with production

Drag-and-drop is the easy half. These are the parts that decide whether what you built still works next week.

Drag a page together

Blocks snap into a real grid — tables, charts, forms, agent chat. What you arrange is what ships, with no separate export step.

Wire the workflow

Triggers, agents, tools, conditions and human approval on one canvas. Steps run in dependency order and stop where they break.

Runs survive restarts

Every step is checkpointed before the next begins. Restart mid-run and it resumes from the last good step instead of starting over.

Approvals hold no process

A run waiting on a person is parked on disk, not pinned to a server. Approve it a minute later or a week later — it continues either way.

Catch the regression

Save any finished run as a test case, then replay the suite after changing a prompt or a model. Passing yesterday and failing today is reported as exactly that.

Answers from your documents

Load the docs that matter and agents retrieve the passages that bear on the question, citing which one they used.

How it works

Describe it, arrange it, run it, ship it

The same declarative spec underneath every step, so nothing you build is trapped in the canvas.

Describe
Say what the product should do. The Copilot proposes a whole workspace — pages, agents, workflow, data — and validates it before anything is replaced.
Arrange
Drag the page into shape and wire the workflow. Every change is the same declarative spec underneath, so nothing is trapped in the canvas.
Run
Execute against live APIs with your own keys. Watch each step's latency, tokens and real cost as it happens.
Ship
Publish to a URL, public or private. Put it on a schedule and it runs unattended.

A workflow canvas: a trigger feeding market data into two agents, a condition, an alert and a dataset write.

Build the first one in a few minutes

Start from a template that already runs against live public APIs, or describe what you want and let the Copilot draft it.

Open the Studio

Create AI apps.Ship anywhere.